MWC 2026: AI+ and Compute Sovereignty — Who Controls the Intelligence Layer?

AI+ and Compute Sovereignty: Who Controls the Intelligence Layer?

At MWC 2026, the conversation has quietly shifted. Last year, the spotlight was on networks becoming AI-native. This year, under the broader AI+ theme, the real question is no longer how infrastructure adapts to artificial intelligence. It is who controls it.

AI is no longer an application running on top of networks. It is becoming the organizing logic of the digital economy. Models decide. Agents act. Systems optimize autonomously. The intelligence layer is emerging as a strategic asset. And whoever owns that layer owns leverage over industry, institutions, and markets.

This is where compute sovereignty enters the debate.

From Connectivity to Control

Telecommunications once revolved around bandwidth, latency, and coverage. Today, those metrics remain necessary but insufficient. Intelligence now sits above connectivity. Large-scale models process data, generate decisions, and increasingly orchestrate other systems.

The shift is structural. In classical computing, software executed predefined logic. In modern AI systems, behavior emerges from trained parameters. These parameters—billions or trillions of them—encode compressed representations of language, images, patterns, and reasoning processes.

As described in contemporary AI discourse, scaling laws have demonstrated that increasing model size, training data, and compute yields predictable improvements in performance. This empirical regularity has transformed AI development into an industrial-scale activity. Intelligence now correlates strongly with access to compute.

This is not merely a technical observation. It is a geopolitical one.

The Scaling Era and the Capital Barrier

The so-called “scaling era” of AI revealed that intelligence improves smoothly as training compute increases across orders of magnitude. Frontier models require vast clusters of GPUs, specialized accelerators, and enormous energy consumption. Training runs cost tens or hundreds of millions of dollars. The barrier to entry is structural.

The result is concentration. Only a handful of corporations and states can finance the infrastructure required to train frontier models. The intelligence layer becomes vertically integrated: data centers, proprietary models, cloud APIs, enterprise tools, and consumer interfaces form a single stack.

In such a landscape, sovereignty is no longer about spectrum licenses or fiber routes. It is about whether a region can train, host, and iterate its own models—or whether it must rely on external platforms.

Inference Scaling: The Economics of Thinking

While training dominates headlines, inference is becoming the new battleground. Recent advances show that increasing compute during inference—allowing models to “think longer” by generating more reasoning tokens—significantly improves performance.

This introduces a new economic variable: reasoning cost per query.

If intelligence improves with more inference compute, then AI capability becomes tied not only to training resources but to ongoing operational expenditure. Enterprises integrating AI agents into workflows must calculate the marginal cost of every autonomous decision.

Inference scaling shifts AI from a capital expenditure problem to a continuous energy and compute allocation problem. This directly intersects with Europe’s regulatory and energy constraints.

AI as an Agent Architecture

Artificial intelligence is best understood through the lens of intelligent agents: systems that perceive environments and act to achieve objectives. In this framing, AI is not merely predictive text generation. It is a decision-making architecture embedded within economic processes.

When enterprises deploy AI agents, they effectively outsource segments of decision-making. Supply chains, pricing strategies, customer support, logistics optimization—all become partially mediated by models trained elsewhere.

If those models are external and opaque, sovereignty is diluted.

Compute sovereignty therefore concerns more than data localization. It concerns control over the objective functions, reinforcement signals, and optimization regimes that guide AI agents’ behavior.

Europe’s Structural Position

Europe possesses research excellence, regulatory leadership, and industrial depth. However, it lacks hyperscale compute dominance. Frontier model training clusters are concentrated primarily in the United States and parts of Asia.

This asymmetry produces a paradox. European companies may comply with the AI Act while relying on foreign foundation models. Regulatory sovereignty without compute sovereignty becomes partial sovereignty.

MWC 2026 reflects this tension. Discussions around AI+ emphasize integration across sectors—automotive, healthcare, finance, manufacturing. Yet integration presupposes access to models. The question is whether those models are European, open-weight, or proprietary external services.

Edge AI and the Fragmentation of Power

One counterbalance to hyperscale concentration is edge AI. By deploying smaller, optimized models directly on devices, enterprises reduce dependency on centralized inference.

This is not purely a latency play. It is a strategic decentralization move. On-device inference lowers marginal cost, enhances privacy, and redistributes control toward hardware manufacturers and local operators.

However, edge models remain downstream of foundational training ecosystems. Without sovereign training pipelines, edge autonomy remains limited.

Energy as the Hidden Variable

Compute is physical. Data centers require land, cooling, and power. As scaling continues, AI becomes an energy-intensive industry. The compute used to train frontier models has grown exponentially over the past decade.

Europe’s energy policy thus intersects directly with AI competitiveness. Renewable capacity, grid stability, and nuclear policy become determinants of AI independence.

Intelligence is no longer abstract. It is thermodynamic.

Open Weights vs. Closed Platforms

Another axis of sovereignty concerns openness. Open-weight models reduce dependency on API-based access controlled by foreign firms. They allow local fine-tuning, auditing, and integration.

Yet open models still require substantial compute to train. The strategic decision for Europe is whether to prioritize open ecosystems supported by public infrastructure investment, or to negotiate privileged access within foreign ecosystems.

There is no neutral path. The intelligence layer will consolidate around specific stacks.

The Enterprise Dilemma

Enterprises attending MWC face a pragmatic question: build, buy, or hybridize?

Building requires internal AI expertise and compute contracts. Buying accelerates deployment but deepens dependency. Hybrid strategies—fine-tuning open models on proprietary data—offer compromise but remain constrained by upstream architecture.

The more AI agents permeate workflows, the harder it becomes to reverse architectural commitments.

Algorithmic Governance and Strategic Autonomy

As AI systems evolve into multiagent environments—interacting models optimizing across markets—the intelligence layer begins to shape macroeconomic behavior. Decision-making becomes probabilistic, adaptive, and partially opaque.

Governance must therefore extend beyond compliance to architectural literacy. Policymakers need to understand scaling dynamics, inference economics, and agent alignment—not only risk categories.

Compute sovereignty is not isolationism. It is strategic optionality.

Barcelona as a Microcosm

Barcelona, hosting MWC, symbolizes this crossroads. The city combines telecommunications heritage, digital startup culture, and European regulatory context. The question facing Barcelona mirrors Europe’s broader dilemma: will it host intelligence, or merely consume it?

Local data centers, research hubs, and AI startups represent seeds of sovereignty. Yet without sustained capital alignment and energy strategy, these remain fragments within a larger external architecture.

The Long Horizon

If scaling continues, frontier models may approach or surpass human-level competence in broader domains. Whether or not full artificial general intelligence emerges, the trajectory implies increasing economic centrality of AI systems.

In such a future, compute sovereignty determines bargaining power. It shapes industrial competitiveness, labor dynamics, and geopolitical influence.

The intelligence layer is becoming as foundational as electricity once was.

Power Above the Network

MWC began as a showcase of mobile connectivity. In 2026, under the AI+ theme, it reveals a deeper transformation. Networks are necessary. Chips are necessary. But the decisive arena lies above them: the models, the inference pipelines, the training clusters, the objective functions.

Compute sovereignty is not a slogan. It is the recognition that intelligence—industrial, economic, strategic—has become infrastructural.

The question is no longer whether AI will shape the future. It is who will own the layer that shapes it.

La sobirania digital del futur no serà només connectivitat, sinó control sobre la intel·ligència.

Further Reading

DeepMind – Mastering the Game of Go with Deep Neural Networks and Tree Search
A landmark example of combining learning, planning, and search — early foundations of agentic system design.

OpenAI – GPT-4 Technical Report
Technical evidence of scaling-driven capability emergence, the foundation upon which modern agentic systems are built.

McKinsey and Company – Artificial Intelligence Insights
Enterprise-level strategic analysis of generative and autonomous AI adoption.

Stanford HAI – AI Index Report 2024
Data-backed overview of global AI acceleration, investment patterns, and enterprise adoption trends.